# OpenAdapt docs — agents.txt

> Verified last-mile execution for agents. Compile a demonstration into a program an agent can invoke. Healthy runs make no model calls. Uncertainty escalates. Humans audit.

Computer-use agents are the user of OpenAdapt. They are not the executor inside OpenAdapt.

Site-wide machine contract: https://openadapt.ai/agents.txt
Walkthrough: https://docs.openadapt.ai/get-started/
Author a workflow (human authority): https://docs.openadapt.ai/get-started/first-workflow/
Run outcomes: https://docs.openadapt.ai/reference/run-outcomes/

## How to call

```
claude mcp add openadapt -- \
  uvx --from 'openadapt-agent[tutorial]' openadapt-agent \
  serve --allow-run
```

```
python -m pip install --upgrade openadapt
openadapt quickstart
openadapt quickstart --break-it
```

`--allow-run` is an explicit opt-in. Never summarize halt as success.

## Outcome vocabulary

`VERIFIED` | `HALTED` | `RECONCILIATION_REQUIRED` | `REFUSED` | `TIMEOUT` | `ERROR`

`VERIFIED` means the independent check passed. Halt is not success. If delivery is uncertain after an action may have been dispatched, the outcome is `RECONCILIATION_REQUIRED`. Do not blindly retry.

Read `report.json` and the seal. `REPORT.md` is for humans.

## Escalation

Halt packets with typed agent-continue are not shipped this week.

- Missing declared parameter: calling agent, if the schema allows it.
- Retryable transport: agent, inside policy, same idempotency key.
- Identity, effect contradiction, expired policy, novel UI, admission: human.

Do not resolve identity or effect contradictions yourself. Do not fabricate a success path.